Pathological Image Analysis and Applications Based on Deep Learning

Pathological image analysis is extremely important in the medical field. With the fast development of deep learning technology, pathological image analysis has changed from manual annotation to intelligent automation. But intelligent systems used for pathological image analysis still have two main challenges: not having enough annotated data and the problem of clinical interpretability. This article solves the above challenges by proposing an intelligent system for pathological image analysis and survival prediction based on deep learning. We use transfer learning and data augmentation to deal with the problem of not having enough annotated data. To reach clinical interpretability, we do segmentation, classification, detection and prognosis analysis. Besides, we also use ensemble learning and hyperparameter tuning. We use LSTM and MLP models for prognosis prediction. Finally, we make clinical interpretability better by visualizing the model results using Grad-CAM. We do experiments on two public datasets. The results show that this method is better than traditional ones.